BlackRock is marketing more than $12 billion in bonds to finance a roughly 1 gigawatt Meta data center campus in El Paso, Texas, Bloomberg reported July 20. The project, codenamed Sopaipilla, is owned 80 percent by BlackRock funds through its Global Infrastructure Partners and HPS units, with Meta holding 20 percent and serving as tenant and developer. JPMorgan and Morgan Stanley are running the sale, with pricing expected in late July.
Meta gets a gigawatt of capacity, due online in 2028, while carrying only a fifth of the project on its own books. The structure mirrors its roughly $30 billion Louisiana financing with Blue Owl, and it is quietly becoming the standard way hyperscale AI gets funded.
How the structure works
The bonds are issued against BlackRock’s 80 percent stake via a project entity, Sopaipilla Investor LLC, with S&P noting a senior secured tranche around $12.3 billion and total size potentially near $13 billion. Bondholders get long-dated exposure secured by Meta’s lease payments; BlackRock’s infrastructure funds get an equity position in an asset with an investment-grade anchor tenant; Meta gets the capacity without the debt. Everyone in the chain is underwriting one thing: Meta’s willingness to keep paying rent on AI compute for decades.
Who buys this paper, and what they are betting on
The natural buyers are insurers and pension funds reaching for long-dated, investment-grade yield. To them the offer reads like classic infrastructure: a hard asset, a multi-decade lease, an anchor tenant with one of the strongest balance sheets on earth. The bet they are underwriting is not really the data center, it is Meta’s willingness to keep paying rent, which insulates bondholders from the compute cycle right up until the lease ends or the tenant walks.
The open question is what kind of asset a gigawatt campus is decades from now. A toll road does not go obsolete when a chip generation changes; a data center’s terminal value depends on continued demand for what runs inside it. The closest historical rhyme is the 1990s fiber buildout, where bondholders funded capacity years ahead of demand. Much of that debt defaulted, and the fiber still ended up lit, just under new owners at new prices. Whether AI compute follows that script is exactly the debate we laid out in the 2026 bubble evidence, and the physical constraints mapped in memory and megawatts cut both ways: scarcity supports rents, but scarcity is also what overbuilding eventually solves.
There is a quieter consequence too. Fixed-income money arrives with covenants, disclosure requirements and rating-agency scrutiny. As more of the buildout is funded this way, the AI economy inherits the bond market’s discipline, which is a very different governor than venture enthusiasm.
The signal
This is the third distinct financing pattern Meta has used in twelve months, alongside the $50 billion Hyperion buildout and its Blue Owl joint venture, while Amazon simply borrowed $25 billion on its own name. The divergence matters: Amazon puts AI debt on its balance sheet, Meta increasingly does not. As we traced in the AI capital stack, the buildout is being funded less by tech company cash flow and more by bond markets reaching for infrastructure yield. That deepens the capital pool enormously, and it means the AI buildout’s risk now sits partly with fixed-income investors who have never priced a GPU cycle before. The deal has not priced yet; the demand it draws will be the real headline.
AI and business tech news, verified by a physician who reads the filings. One email a week, no noise.
